In late 2025, the U.S. Department of Justice shut down a major China-linked smuggling network that allegedly routed tens of millions of dollars’ worth of H100 and H200 GPUs to China by falsifying documentation and relabeling hardware . In a similar case, DeepSeek was accused of establishing “ghost” data centers in Southeast Asia to pass audits, then shipping GPUs onward .
These cases all highlight the difficulty of enforcing controls once hardware leaves Nvidia’s hands. Export rules are primarily enforced at the point of sale and shipment and rely heavily on declarations of end use and on downstream compliance by resellers and customers. When demand is strong enough, the incentives to circumvent those declarations multiply — and China’s appetite for AI compute remains enormous.
Domestic alternatives, including Huawei’s Ascend accelerators, have improved but still lag Nvidia in software maturity and ecosystem support. Even Chinese firms that publicly promote local silicon often rely on Nvidia hardware for training large models or running advanced inference workloads. That persistent demand has created a shadow market willing to pay significant premiums for restricted GPUs.
From Washington’s perspective, export controls are intended to impose real friction on China’s AI development by constraining access to the highest-end compute. The logic is that advanced model training scales with available GPU throughput, memory bandwidth, and interconnect performance. Denying that hardware should slow progress in both commercial and military AI applications.
The problem is that for China, which is so far behind the West in real terms, partial access is still extremely meaningful. Even a modest influx of smuggled or otherwise indirectly routed GPUs can support the likes of research projects and inference deployments. While it’s estimated that illicit imports account for only a fraction of China’s total compute capacity, even marginal gains can make a big difference at the frontier of model development.
At the same time, aggressive controls carry trade-offs. They push Chinese companies to accelerate domestic chip development, fragment global supply chains, and incentivize exactly the kinds of gray-market behavior now under investigation. They also place companies like Nvidia in a difficult position, caught between compliance obligations and the commercial reality that much of their addressable market lies outside the U.S. and its closest allies.
In early 2025, the Commerce Department expanded controls to cover not just hardware but also certain AI model weights, while creating new licensing frameworks for trusted partners and data center operators.
While some of these controls have been relaxed by President Trump, with H200 sales now permitted to vetted Chinese customers subject to a 25% import duty, the back-and-forth indicates clear uncertainty among decision-makers about what level of restriction, if any, will achieve the desired outcome. Indeed, it’s difficult to argue that allowing the H200, some of the most advanced silicon made by Nvidia, does anything to serve the U.S. goal of hindering China.
Whether or not investigators ultimately find evidence that Megaspeed violated export laws, there’s a structural weakness in the current system. Export controls assume that intermediaries can be trusted to enforce end-use restrictions at scale, across borders, and over time. As long as global demand for Nvidia-class AI compute outpaces legal supply to China, however, pressure will build on the seams of the system.
Follow Tom's Hardware on Google News , or add us as a preferred source , to get our latest news, analysis, & reviews in your feeds.
Luke James Social Links Navigation Contributor Luke James is a freelance writer and journalist. Although his background is in legal, he has a personal interest in all things tech, especially hardware and microelectronics, and anything regulatory.
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://www.tomshardware.com/pc-components/gpus/SPONSORED_LINK_URL
- https://www.tomshardware.com/pc-components/gpus/nvidias-biggest-sea-customer-exposes-the-limits-of-us-ai-export-controls#main
- https://www.tomshardware.com
- ASRock X870 Taichi Creator Review: A Taichi for Creatives
- Intel's upcoming Xeon Granite Rapids workstation lineup leaks, poised to challenge AMD Threadripper with $8,300 86-core flagship — retailer lists prices ahead o
- Into the Omniverse: OpenUSD and NVIDIA Halos Accelerate Safety for Robotaxis, Physical AI Systems
- UC San Diego Lab Advances Generative AI Research With NVIDIA DGX B200 System
- NVIDIA Awards up to $60,000 Research Fellowships to PhD Students
Informational only. No financial advice. Do your own research.